Apple Just Locked Down AI Agents. Your Content Still Needs to Be Readable by Them
Two Signals, One Week
Two things happened the same week and almost nobody connected them.
Apple tightened Full Disk Access on macOS on October 2, 2026. The update requires very explicit user action before an app gets deep access to your files, and Apple's own developer notes cite AI agent risk as the reason. Autonomous agents are capable enough now that Apple decided the old permission model wasn't safe anymore. That's a device-side story. It's about what software can touch on your Mac without you saying yes first.
The same week, a 47-point AEO checklist got updated on October 1. Citelity's version leans hard into JSON-LD schema, Person author objects, dated publishing fields, and dedicated source pages with original data. That's a content-side story. It's about what AI models are willing to cite when they're answering a question for someone.
Put those next to each other and the pattern is not subtle. Agents are getting restricted on the device because they can do more than they used to. At the same time, the content those agents pull from is getting scrutinized harder before it gets quoted or recommended. Access is tightening on one end. Sourcing standards are tightening on the other.
If you run a website and you've been treating AEO like a vague future problem, this is the week that makes it concrete. The agents are more capable and more supervised. The content needs to be more verifiable to earn their attention.
Schema That Actually Gets Read
Most sites still run FAQ schema on everything. Product pages, blog posts, the homepage. It made sense a few years back when FAQPage rich results actually showed up in search. That mechanism got retired for a lot of use cases back in 2023, and the 2026 checklists finally caught up to that reality. Stacking FAQ markup on a page today mostly produces code nobody reads.
What citelity's October 1 checklist actually weights is Article schema, done fully. Not a stub with a headline and a date. A Person object for the author, with a name attached to a real identity rather than "staff" or nothing at all. An Organization object tying the piece back to the publisher. A datePublished field that's accurate and present, because an AI model deciding whether to cite something fresh checks that field before it checks anything else. BreadcrumbList rounds it out, giving the model a sense of where this page sits in your site's structure.
None of this is exotic. You can validate all four with Google's Rich Results Test in a few minutes. The work isn't technical difficulty, it's discipline. Go through your last twenty posts and check how many have a real author object instead of a placeholder. Most sites fail that check before they even get to the FAQ question. Fix the boring fields first.
Build a Source Page, Not Another Blog Post
a blog post optimized within an inch of its life still competes with every other blog post saying roughly the same thing. A source page doesn't compete with anything, because nobody else has your numbers.
The 2026 checklists are blunt about this. Original data with a named primary source for every stat is now a requirement, not a nice-to-have. Anonymous claims, "studies show," "research indicates," that kind of hand-waving gets filtered out before a model ever considers citing it. If you can't point to who ran the number, the model has no reason to trust it over a competitor's version of the same claim.
So build the page differently from the start. Pick one data point your business actually owns, response times, conversion rates on a specific workflow, time saved on a repeatable task, and attach a name to where it came from. Internal survey, a tool's own usage logs, a named report. Structure the page as a reference, not a narrative. No scene-setting, no "in today's landscape." Just the number, the method, the source, stated plainly near the top.
Think of it as a page built to be pointed at, not scrolled through. Scrolling is for blog posts. Citation is for pages that answer one question completely and tell you exactly where the answer came from.
Write the 40-120 Word Block They Can Lift
Here's a test worth running on your own site right now. Pull up any paragraph from your last post and ask whether it would survive being lifted out, dropped into an AI answer, with zero surrounding context. Most paragraphs fail immediately because they lean on "as mentioned above" or "building on this point," phrases that only make sense inside the flow of the full article. A model quoting you doesn't carry your flow with it. It carries the sentence.
The fix is writing chunks between 40 and 120 words that answer one specific question completely, in order, with no dependency on what came before. Lead with the answer, not the setup. If someone asks what your refund policy covers, the first sentence should state the policy. The explanation of why you built it that way can come after, but it's optional context, not load-bearing.
Write five or six of these per page, each tied to a question a real customer would type into a chat window. Then validate the schema wrapping them through Google's Rich Results Test before you publish. It takes two minutes and tells you whether the Article markup you just added actually parses the way you think it does.